espnet.nets.pytorch_backend.transducer.rnn_encoder.Encoder

source

class espnet.nets.pytorch_backend.transducer.rnn_encoder.Encoder(idim: int, etype: str, elayers: int, eunits: int, eprojs: int, subsample: ndarray, dropout_rate: float = 0.0, aux_enc_output_layers: List = [])

Bases: Module

Encoder module.

• Parameters:
  • idim – Input dimension.
  • etype – Encoder units type.
  • elayers – Number of encoder layers.
  • eunits – Number of encoder units per layer.
  • eprojs – Number of projection units per layer.
  • subsample – Subsampling rate per layer.
  • dropout_rate – Dropout rate for encoder layers.
  • intermediate_encoder_layers – Layer IDs for auxiliary encoder output sequences.

Initialize Encoder module.

forward(feats: Tensor, feats_len: Tensor, prev_states: List[Tensor] | None = None)

Forward encoder.

• Parameters:

  • feats – Feature sequences. (B, F, D_feats)
  • feats_len – Feature sequences lengths. (B,)
  • prev_states – Previous encoder hidden states. [N x (B, T, D_enc)]

• Returns: Encoder output sequences. (B, T, D_enc) : with or without encoder intermediate output sequences. ((B, T, D_enc), [N x (B, T, D_enc)])

  enc_out_len: Encoder output sequences lengths. (B,) current_states: Encoder hidden states. [N x (B, T, D_enc)]

• Return type: enc_out